COMMUNICATION OF DEPENDENT MESSAGES OVER COMPOUND CHANNELS

Abstract

In the communication of multimedia content, certain dependency structure often exists among the source-coded messages by different source coding techniques, where by “dependency” we mean the dependent contributions of the messages to the overall reconstruction quality. Motivated by such notion of dependency, this thesis considers the problem of communicating dependent source-coded messages over compound channels, which include the attractive wireless channels and packet-loss channels. In this thesis we propose a novel framework to model arbitrary dependency structure among sourcecoded messages from the source-reconstruction perspectives, and formulate the problem of communicating such messages over compound channels as the problem of maximizing the average utility at the receiver. Over discrete memoryless channels (DMC), we derive the expression of maximal achievable utility, which appears to be governed by the channel coding theorem. Over degraded compound channels, we study analytically the maximal achievable utility by superposition codes. To achieve the maximal utility, the encoder chooses the best sub-chain in the utility graph and encodes it using the best superposition code. For the case of two source-coded messages, we show that the maximal utility achieved by superposition codes is the maximum among all coding schemes. Since in practice layered codes (LC) and multiple description codes (MDC)